The classification of human hand gestures has gained widespread recognition as a natural and powerful way to interact intuitively and efficiently with computers. Specifically, this approach has facilitated the development of a number of important applications in the medical training field, specially as a way to objectively evaluate surgical tasks of novices compared to an expert surgeon. In this paper, 3D medical gestures, acquired by an instrumented laparoscopic forceps, were classified using Long Short Term Memory (LSTM) recurrent neural networks (RNN). Recognition results are based on gesture dynamics and a comparison of gesture trajectories between novices to an expert motion are presented. Using LSTM RNN, we were able to achieve a recognition rate of 99.1%.